While true, what surprised me is that Knuth's books are much more readable than you might expect from their reputation. TAOCP is a far cry from dry terse impenetrable tomes like Rudin.
One further than this: every time --- every time --- I sit down with TAOCP, even on stuff I assume is going to be basic, I learn something awesome. It is one of the densest amounts of awesome per page of any book I've ever seen.
I think it's the sheer, awe-inspiring _quantity_ of high-level material that's hard to take. I've read parts of of TAOCP very deeply - implementing things, trying exercises, etc. and it is rewarding/exhausting. I doubt that I've worked through much more than 70-80 pages in this fashion from all of the books.
Given enough time, one could probably do this for the whole series to date, but I'd like to at least be working part-time before attempting it.
I never believed the story. It seemed out-of-character vulgar and hostile for Knuth.
The only Rudin book I've worked through is Real and Complex Analysis, and it is very dry indeed.
He also seems to favor some less intuitive proofs sometimes, which while correct and complete, leave the reader in the dark as to the intention behind it (the proof for the Lebesgue-Radon-Nikodym theorem given is an example of that.)
I found that the introduction to MIX (MMIX in the latest editions) at the beginning was hard to get through. I pick up TAOCP as a reference once in a while, and I agree that later parts of it are much more readable.
While true, what surprised me is that Knuth's books are much more readable than you might expect from their reputation. TAOCP is a far cry from dry terse impenetrable tomes like Rudin.